EmoXract: Domain independent emotion mining model for unstructured data

Abstract

Emotion plays an important role in human computer interaction to give a human like feel. To acknowledge the importance of emotions in an artificial agent, we propose a domain independent emotion mining model (EmoXract) which extracts emotions from an unstructured data. The emotion is extracted at sentence level based upon the contextual information. Basically, we have used two corpuses: WordNet dictionary and WordNet-Affect dictionary. WordNet dictionary is used for the creation of synonyms and stemmed words. WordNet-Affect dictionary is used to establish a weighted relationship between each word to every primary emotion. Various modules adopted in the model are converter, tokenizer, creating synsets and stemmed words, assigning weights, heuristics rules, calculating net weight and sentence level emotion extraction. We have also designed a self-learning dictionary which self-updates the new word, its synonym and stemmed words with the same weight in accordance to its already existing synonym. Finally the model is simulated for a test data of more than 500 sentences, selected from different domains to validate the proposed design.